Talk:Hard disk drive/Archive 13

Archive 10

Archive 11

Archive 12

Archive 13

Archive 14

Archive 15

→

Archive 20

2016 desktop capacity revised forecast

Section title is "Future Development"

Original text: HDD areal densities have shown a long term compound annual growth rate not substantively different from Moore's Law, most recently in the range of 20–25% annually, with desktop 3.5" drives estimated to hit 12 TB around 2016.(2012) p. 8

==> The orginal forecast of 2012 will be revised, since forecasts are more accurate when revised often. The original chart on page 8 has 12 GB for 2016; 2 GB for 2011. This is a 40% CAGR, with 6 GB for 2014??? Desktops now (2014) have 1 TB/disk according to Coughlin (2014) ref: http://www.forbes.com/sites/tomcoughlin/2014/04/07/new-areal-density-point-for-cloud-storage-hdds/

HDD areal density has shown a long term compound annual growth rate of 25-100% annually, not substantively different from Moore's Law, and most recently in the range of 10–25% annually.

(2012) p. 8

[(2013) https://www.dtc.umn.edu/resources/bd2013_anderson.pdf]

However beginning around 2010, the superparamagnetic limit contrained the rate of progress. [(2013) https://www.dtc.umn.edu/resources/bd2013_anderson.pdf]

Consumer grade desktops in 2014 had 1 TB per 3.5-inch disk or 4 TB per drive. [(April 2014) http://www.forbes.com/sites/tomcoughlin/2014/04/07/new-areal-density-point-for-cloud-storage-hdds/]

With areal density growth of 20%, this would be estimated to hit 1.4 TB per disk or 6 TB per drive around 2016. ==== 71.128.35.13 (talk) 23:20, 23 May 2014 (UTC)

I believe the 4 TB desktop HDD was introduced as early as Jan 2013 which suggests the base year for your calculation is wrong. Given 6 TB Enterprise HDDs are readily available this year u might want to reconsider your extrapolation. Tom94022 (talk) 16:56, 2 June 2014 (UTC)

Are you showing 1.0 TB per consumer grade (minimum cost per bit) disk in stores (not just on paper) at start-of-2013, using an assumed 20% CAGR? That implies mid-2014 would have 1.3 TB / consumer grade disk, and mid-2016 would have 1.9 TB / consumer grade disk. But rather than 1.3 TB, I still see 1.0 TB per consumer grade disk in stores now. So from where I sit, density CAGR has been 0% since early 2013.

The 1.9 TB / consumer grade disk scenario in mid-2016 would require, starting from the actual density observed now in mid-2014 of 1.0 TB / consumer grade disk, (1.9 / 1.0) ^ (1 / 2 years) -1 = 40% per year growth rate from here on out. This is not likely.

So my confidence in consumer grade CAGR even reaching as high as 20% per year would be reduced if areal density at the consumer level is stagnant. It could be less than 20% per year now.71.128.35.13 (talk) 03:41, 3 June 2014 (UTC)

The Seagate Barracuda 7200.15 at 4TB was commercially available in early 2013 - do your own research to figure out what was its date and if it was first. AD advance is not continuous, it a discrete series moving in discrete steps, which are fewer now that there are only three companies. There is no growth until the next new product announcement - this is NOT zero or "stagnant" grown, just no data. Tom94022 (talk) 05:31, 3 June 2014 (UTC)

No density data, but the price data are indicative and supportive. Retail prices in the last year have improved (dropped) by around 20% per year according to McCallum(2014),
http://www.jcmit.com/disk2014.htm
http://www.jcmit.com/diskprice.htm
and areal density moves together with price (and vice-versa) over the long term. Consumer grade density has not changed in the last 18 months, because suppliers can't change areal density on their manufacturing lines daily as they can on prices at retail. Even so, the areal density growth forecast of <20% per year is in the same ballpark as the approximately 20% price improvement observed over the past 12 months.71.128.35.13 (talk) 20:51, 3 June 2014 (UTC)

Another goalpost change? The section talks about bytes/3.5-in disk which is directly related to AD - cost/bit is something else and it is not uncommon for the first product at a new AD to be announced at a high price/bit and then price decreases as yields and volumes improve. At the MSST2014 conference today a Seagate representative stated that a new higher density desktop type product will be announced this year using shingled write - a 6 TB drive on 4 disks would be right in line with 20%/year from early 2013 and two years before your prediction. BTW, I am not sure any original research prediction belongs in Wikipedia but I want to think about it before I remove it. Tom94022 (talk) 23:05, 3 June 2014 (UTC)

Did the original article text, forecasting 12 TB in 2016, not need any references, or was that original research? The previous forecast (based on 40% CAGR) has no references, so I was slow to add references when I updated it.

The 20% forecast isn't mine. The references for the <20% forecast are as follows: a Stanford university data storage researcher and other researchers (most not on the industry payroll) published at an UNESCO conference; an industry roadmap quoted by an industry analyst, and a senior Seagate technical manager. I will add the three references.

Were you at the 2012 or 2013 conference when Seagate talked about shingled recording soon? I don't see any shingled products on the store shelves that break below existing price floors. Today's announcement may or may not have an effect on minimum-cost-per-bit consumer-grade drives in the stores. I'm not talking about enterprise products, vaporware or specialty goods for particular market segments. If the 1.4 TB per disk product achieves a new cost minima (this year, in the stores, consumer grade) then it would count. The 20% forecast would put 1.3 TB per disk HDD in the stores today, based on a January 2012 start date. I'm seeing 1.0 TB per disk now. The 20% CAGR density forecast stands as the best substantiated, best referenced. — Preceding unsigned comment added by 71.128.35.13 (talk) 00:47, 4 June 2014 (UTC)

I deleted it. It violates WP:NOTCRYSTAL and WP:SYNTH. It doesn't matter what anyone here thinks would be "right in line" with anything. If you can find a prediction for a specific capacity in a specific year, you can quote or paraphrase that and cite it. If you find a prediction that some factor like areal density increase rate will continue, you can quote or paraphrase that and cite it. But you cannot do the arithmetic and thereby predict a specific capacity in a specific year—not with Wikipedia's voice. The above-cited guidelines are very clear. There really isn't any doubt or ambiguity about it. Jeh (talk) 00:52, 4 June 2014 (UTC)

WP:EW Discussion on talk page on the part of Tom94022 is suggested prior to initiating conflict.

Original text: "for 2011 to 2016, [Anderson(2013)] [isuppli(2012)] areal density compound annual growth will be in the range of 19 to 40% per year." The earlier version based on these references, "<20%/yr", was dismissed as too much information.

Neither reference supports the forecast of 40% per year: Anderson(2013) of Seagate has 8-12% per year, and the isuppli(2012) industry roadmap has 19% per year. Could you provide an properly-sourced, recent reference for 40% per year? 71.128.35.13 (talk) 22:53, 5 June 2014 (UTC)

Anderson shows 40% on page 2 for the relevant period. Coughlin shows 25%. iSuppli shows 19%. UNESCO is not a reliable source since it depends upon iSuppli which apparently does not use the word roadmap but even it did, iSuppli is one analyst among many and not the source of an industry roadmap. The Anderson 8-12 percent is recent history and not a projection. Accordingly I am reverting. Tom94022 (talk) 23:37, 5 June 2014 (UTC)

I honestly think you two are trying way too hard at the "my source is better than your source" game... or at least, at trying to come up with a single defensible number, and dismissing the number derived by the other person. Analyzing the sources' sources, reasoning, use (or not) of specific words like "roadmap", etc., is WP:SYNTH. From WP:NOTCRYSTAL:

"Predictions, speculation, forecasts and theories stated by reliable, expert sources or recognized entities in a field may be included, though editors should be aware of creating undue bias to any specific point-of-view."

So: When different sources make different predictions, the most neutral thing is to just cite each of them, preceding all with an intro sentence such as "Various industry analysts have published differing predictions for future improvements." Then you can both walk away. Jeh (talk) 01:35, 6 June 2014 (UTC)

Jeh, the current construction almost verbatim meets your neutral proposal. The range "19 to 40% per year", bounds three reliable sources, Anderson (40%), Coughlin (25%) and iSuppli May 2012 (19%). UNESCO is at best a tertiary source since the sentence used is directly attributed to iSuppli May 2012; UNESCO appears to be an unreliable source since the quoted sentence is not supported by the iSuppli report as it is summarized in the link. The iSuppli report is not public so we don't know exactly what is says but that really doesn't matter because the lower bound is either 20% as in UNESCO or 19% as in ISuppli May 2012. The quoted UNESCO statement "The current roadmap predicts no more than a 20%/yr improvement in bit density" is clearly at odds with Anderson and Coughlin and misstates iSuppli in several ways so it really doesn't belong here. Tom94022 (talk) 03:46, 6 June 2014 (UTC)

Jeh, the following genuinely and verifiably complies with your neutral proposal, without misrepresenting Anderson(2013). The 40% mentioned is plainly historical as pages 3 and 4 indicate clearly, and quoting verbatim, “CAGR slowing from historical 40+% down to ~8-12%”. The updated and revised Coughlin (2013) supersedes Coughlin (2012) which had 25%.

"Various industry analysts have published differing assessments of HDD areal density compound annual growth rate: isuppli (2012) forecasts 19% for 2011-2016; Seagate (2013) has 8-12% for c. 2013; analyst Tom Coughlin (2013) has 0% (flat) for 2011-2013 and the longer term cumulative average is currently (c. 2013) less than 20%; and, John Rydning of IDC (2013) has less than 20% for c. 2013."71.128.35.13 (talk) 19:00, 6 June 2014 (UTC)

Tom94022, you write (03:46, 6 June 2014) that “'The current roadmap predicts no more than a 20%/yr improvement in bit density' is clearly at odds with Anderson and Coughlin”. But on the contrary, Anderson(2013) actually says “CAGR slowing from historical 40+% down to ~8-12%”. The Coughlin(2012) reference is outdated, assuming that you weren't aware of Coughlin's revised 2013 forecast. Coughlin (September 2013) now writes “the average annual cumulative annual growth rate (CAGR) in areal density is currently less than 20%.” Could you provide a properly-sourced, recent reference for 40% per year? With no supporting reference, the figure of 40% per year in the article is wrong and misleading. 71.128.35.13 (talk) 18:22, 7 June 2014 (UTC)

The sentence is prospective covering the period 2011 thru 2016, all of your cites above are retrospective with a cutoff in 2013 and therefore are irrelevant. Anderson is prospective thru 2020 and clearly shows on page 2 his forecast of 40%/yr CAGR thru at least 2019. He does note that retrospectively it has been in the 8-12% range but that doesn't have much to do with his prospective forecast (if u look carefully at his graph, u can see the slowing down in the transition from PW&GMR to HAMR). Likewise the new Coughlin doesn't replace the original Coughlin forecast but mere notes that "the average annual cumulative annual growth rate (CAGR) in areal density is currently less than 20%." (emphasis added) and that "It has been more than two years since there was an increase in the areal density of HDDs." He does note that this has happened in the past so there is no reason to assume that this pause is any different than one in the past. Actually his whole Forbes article is all about "research to create higher areal densities continues." For the record he is wrong about the latest increase in AD since according to my original research the latest increase in areal density was Seagate's LaCie announcement on Feb 20, 2014 of a 5 TB HDD at an AD of 826 Mb/sqin up from Toshiba's 744 on Aug 2, 2011 - indeed a long pause but not particularly relevant to this discussion since as it turns out the Toshiba increase was particularly large. After all the above we still are left with 19% (iSuppli), 25% (Coughlin) and 40% (Anderson) as reputable sources of forecasts thru 2016 and beyond. Tom94022 (talk) 22:18, 8 June 2014 (UTC)

A forecast of "less than 20" describes a ceiling, not a bounded range. It sets no floor. The 8-25% range of current assessment is obviously at odds with the 0% figure of Coughlin (2013). Furthermore, the cite-them-all neutral proposal should not be neglected, brushed aside or bluffed with endpoints that do not name ALL the sources and selectively exclude some sources. The following neutral framework cites all assessments in chronological order:
"Various industry analysts have published differing assessments of current HDD areal density compound annual growth rate: Marchon et al. (2013) have 20-30% for the early 2010s; Seagate (2013) has 8-12% for c. 2013; analyst Tom Coughlin (2013) has 0% (flat) for 2011-2013 and the longer term cumulative average is currently (through 2013) less than 20%; John Rydning of IDC (2013) has less than 20% for c. 2013; and Tom94022 (mid-2014) (of course this original research can't be used in the wikipedia article) has 4% for August 2011 to February 2014."

This statement cites all the forecasts:
"Looking at the growth rate going forward, various industry analysts have published differing forecasts: isuppli (2012) forecasts 19% for 2011-2016; Coughlin (2012) has 25% for 2011 to 2016, Marchon et al. (2013) have 25% for 2010-2020, and Seagate (2013) has HAMR/Extension attaining 20-40% for 2015 to 2020."

To reiterate: cite them ALL. 71.128.35.13 (talk) 20:59, 9 June 2014 (UTC)

I'm surprised with all your research you missed Coughlin's April 7, 2014 identification of the recent 25% increase in areal density noting that this is the first increase since Q3 2011. In a discreet event series such as areal density the absence of an event is not data and that is all Coughlin's 0% is, the absence of an event. The way u seem to want it is until the day before an announcement the rate is 0% and on the day of this 25% announcement the CAGR is 2E37%. I am also surprised at your continued reliance on UNESCO which is clearly a mis-statement of iSuppli; the forecast of 19% by iSuppli is not a "ceiling of 20%" as restated by UNESCO.
Regardless, as far as your complete statement goes, I don't see any disagreement now as to the facts but only its composition. IMO, it is [[WP:TMI| to much information ] to much information to include all of the forecasters and their forecasts in the sentence, the existing sentence reads better, bounds all the sources and can include all of them as footnotes. Tom94022 (talk) 00:11, 10 June 2014 (UTC)

I don't research, however, I read reports from those who do. I did have to research the meaning of TMI. Perhaps wikipedia has cast “too much information” into the dustbin of history. Is it no longer useful? Wikipedia:Articles for deletion/Too Much Information Some other non-obsolete regulation may serve to gum up this discussion, but not this one.

At any rate, could you discuss ASAP with Jeh your concern over the awkwardness, inconvenience and bloat of listing out all the assessments and forecasts? This “TMI” concern went unmentioned when you responded to Jeh three days ago. Jeh wrote as follows: “So: When different sources make different predictions, the most neutral thing is to just cite each of them, preceding all with an intro sentence such as "Various industry analysts have published differing predictions for future improvements." Then you can both walk away. Jeh (talk) 01:35, 6 June 2014 (UTC)”

I hold fast, very fast, to Jeh's cite-them-all neutral proposal. The framework is already here. There isn't, in my view, sufficient justification (like bloat) to deviate from the purity and simplicity of cite-them-all neutrality. After several decades of rapid areal density growth, the ground now has become muddy. Fly heights have dropped; asperities threaten. Simplifying the outlook through the lens of Wikipedia won't work.

According to the new Coughlin (2014) reference you cite, 2014 April is 1.25 times the 2013Q2 value, which itself is 1.07 times the 2011Q3 value. In this case, averaged over the entire time frame under observation, analyst Tom Coughlin (2014) has 12% for 2011Q3-2014Q2. The Coughlin(2013) figures were 0% shorter term and “less than 20%” longer term. So, Coughlin's views have remained consistent: nothing to see here; move along.

BTW in fact I haven't continued to rely on UNESCO for current assessments or forecasts, only because other assessments like isuppli are so abundant. Still, UNESCO is far and away more authoritative, less subject to monetary influence, broader in perspective and far-sighted than the industry sponsored press releases which are the bread and butter, the Oracle of Delphi, to some analysts in this corner of the technology industry. UNESCO's findings deserve note elsewhere in the HDD article. They've raised my eyebrows. Instead of attacking their report, which isn't even relevant to this discussion as it is not present in either assessments or forecasts, it would be more relevent to find more and newer sources as you did with Coughlin(2014). Huge progress is now evidenced by your statement that you “don't see any disagreement now as to the facts”. The current assessments now stand as follows:
"Various industry analysts have published differing assessments of current HDD areal density compound annual growth rate: Marchon et al. (2013) have 20-30% for the early 2010s; Seagate(2013) has 8-12% for c. 2013; John Rydning of IDC (2013) has less than 20% for c. 2013; and, analyst Tom Coughlin (2014) has 12% for 2011Q3 to April 2014 comprised of two phases, 28% after 2013Q2 and 4% before."

Note the bounds on these forecasts are quite wide and technically unbounded: Marchon et al. has 30% at the high end, but this is not very credible in light of the other assessments, while John Rydning of IDC (2013) has no lower bound. Therefore, the mathematics and the semantics say there is no floor; the rock bottom is not 8%, and the "range" is really a ceiling.
Density forecasts, as opposed to assessments, remain unchanged.

Each of these cites is not a forecast but instead a report on actual results and as such do not belong in the referenced sentence (emphasis added):

• Marchon et al. (2013) have 20-30% for the early 2010s ;
• Seagate(2013) has 8-12% for c. 2013;
• John Rydning of IDC (2013) has less than 20% for c. 2013;
• analyst Tom Coughlin (2014) has 12% for 2011Q3 to April 2014 comprised of two phases, 28% after 2013Q2 and 4% before."

All of these cites are retrospective, not forecasts. In the end we are left with three forecasts which are subsumed with the range 19-40% representing all of the forecasts. The detail u want belongs in a footnote at best. JEH did not say literally all of the information on all of the forecasts so the current construction is IMO fully consistent with a neutral presentation of forecasts. My question to you is what do we add to the sentence other than complexity by adding all this detail. Tom94022 (talk) 06:29, 10 June 2014 (UTC)

We get accuracy. Complexity is a small price to pay for accuracy. Otherwise, it would be simply wrong. This subject requires more than one sentence for accuracy and completeness. Jeh wrote as follows:

So: When different sources make different predictions, the most neutral thing is to just cite each of them, preceding all with an intro sentence such as "Various industry analysts have published differing predictions for future improvements." Then you can both walk away. Jeh (talk) 01:35, 6 June 2014 (UTC)

The situation here is the same: different sources making different current assessments. The solution is the same: When different sources make different assessments, the most neutral thing is to just cite each of them, preceding all with an intro sentence such as, various industry analysts have published differing assessments of the current rate of improvement.

• Marchon(2013b) does indeed forecast HDD areal density, 25% CGR for 2010-2020; see Figure 1. The Marchon(2013b) forecast should not be lumped together with the Marchon(2013a) current assessment, even though they share similar numerical values.
• Seagate (2013b) does in fact forecast HAMR/Extension attaining 20-40% for 2015 to 2020 (see slide 11), while Seagate(2013a) currently assesses 8-12% c. 2013.

Furthermore, Wikipedia is misquoting Seagate(2013b) by selecting only 40% when Seagate's full range is actually 20-40%. This is an error, and I'll fix the article.

The following belongs in a separate sentence discussing current assessments, not forecasts:

Various industry analysts have published differing assessments of current HDD areal density compound annual growth rate: Marchon et al. saw 20-30% during the early 2010s; Seagate noted 8-12% c. 2013; John Rydning of IDC described less than 20% c. 2013; and analyst Tom Coughlin observed 12% from 2011Q3 to April 2014.

The following belongs in a separate sentence discussing forecasts, not current assessments:

Looking at the HDD areal density compound annual growth rate going forward, various industry analysts have published differing forecasts: isuppli (2012) expects 19% for 2011-2016; Coughlin (2012) forecasts 25% for 2011 to 2016, Marchon et al. (2013) predict 25% for 2010-2020, and Seagate (2013) anticipates HAMR/Extension attaining 20-40% for 2015 to 2020.

FWIW, I disagree with both your proposed sentences for different reasons and will get around to responding in due course. BTW, it is not an error to say a the range of forecasts is 19-40 which would include one with range of 20-40. Tom94022 (talk) 23:44, 23 June 2014 (UTC)

CITATIONS (added to make linking a bit easier)